Conventional regression methods for modelling the number of road accidents at the district or county level treat the numbers of road accidents for different observational units as independent. Such models seldom take account of any spatial effects, and therefore may not fully account for the spatial variation in the response variable - residuals from the model may be spatially autocorrelated. This paper shows how the hierarchical Bayesian model used in disease mapping can also be used to model road accidents aggregated at the local authority level in England. Models with and without spatial effects are compared. Various neighbourhood relationships are considered, based on both geographical proximity and the structure of the road network. Moran's I standard measure of spatial autocorrelation, is used to illustrate how the inclusion of spatial effects reduces the residual spatial autocorrelation, and to investigate different spatial weighting schemes. Data obtained from previous research studies are used to fit the models. Results from the models are plotted on a map of England at the desired level created in ArcView3.(A) Reprinted with permission from Elsevier. For the covering abstract see ITRD E134766.
Abstract